Research On Feature Recognition And Recommendation Algorithm Based On Feature Association

With the development of the Internet of things and the advent of the era of big data,the explosive growth of data has led to problems such as information overload,and the traditional recommendation system has gradually transformed into a personalized recommendation system.The personalized recommendation algorithms provides information filtering and referral services by building user portraits and predicting user behavior.Recommendation system under the background of big data,the data structure of domain recommendation technology is increasingly complex,presenting new features such as massive heterogeneous data,missing data features,and data feature anomalies and data feature associations.These features put forward new requirements and challenges for the recommendation algorithm from the aspects of problem size,degree of missing feature,abnormal feature state and association relationship.Therefore,the research on prediction and recommendation algorithms based on feature association are carries out in this paper.Mainly includes the following aspects:(1)Heuristic mining algorithm and feature matching algorithm for classification association rules of feature associations.Based on the implicit association relationship of massive data,this paper mainly focuses on the implicit classification association rules of the data itself.By introducing classification and continuous data feature properties and discretize them,and extend the binary representation of data features to ensure the diversity of data feature properties.In order to explore some feature associations in the data,the heuristic feature mining method based on minimum support is studied,and the frequent features of the associated features are found and the optimal feature subset is constructed.Based on the frequent items of data features,the heuristic classification association rule mining algorithm based on minimum confidence is studied.For different scenario pattern,the recommendation algorithm based on feature matching of classification association rules is carried out.The experimental verification and analysis of the health medical scenario experimental data in the machine learning repository were carried out to verify the effectiveness of the proposed algorithm.(2)Implicit feedback feature recognition and prediction algorithm.Aiming at the problem of data feature sparsity and missing of application domain,the systematic feature recognition and prediction classification solution of data feature missing problem in the research domain are studied.On the basis of systematically analyzing the missing data features in the domain,the collaborative filtering feature recognition method based on weighted users is studied.Through the transformation from supervised learning to unsupervised learning,the feature recognition method of implicit association between feature properties in recommendation system is studied.The implicit feedback collaborative filtering feature recognition and prediction algorithm based on implicit feature extraction is studied.The effectiveness of the algorithm is verified by randomly generating missing data feature to simulate the real environment data.The experimental verification and analysis of the health medical scenario experimental data in the machine learning repository were carried out to prove the predictive accuracy and effectiveness of the proposed algorithm.(3)Data anomaly feature recognition and prediction algorithm.In view of the limitations only focus on the discrete data feature properties,the feature recognition algorithm based on the interdependence association of continuous data sequences is studied to recognize anomaly features and prediction.The feature extraction method of continuous time series data based on deep learning network model is studies.Through dimensionality reduction processing of complex graph pattern data and time-frequency sequence data analysis,a deeply data time-sequence dependency and anomaly feature recognition model are formed to improve the effectiveness of predicted result.The experimental verification and analysis of electroencephalogram health medical scenario experimental data were carried out to prove the predictive accuracy and effectiveness of the proposed algorithm.(4)Domain-oriented cascading and weighted hybrid personalized recommendation method.In view of recommendation requirements of specific domains,the hybrid recommendation methods under different scenarios modes are studied,and the research problem of the domain is abstracted into the personalized recommendation problem of recommendation objects for the ontology.The user feature properties model portrait is constructed,and similar users are found by the similar user discovery algorithm based on classification tree and content similarity.The weighted calculation based on the feature matching algorithm of the association rule is used to obtain the recommended scheme.For the cold start problem of recommendation algorithm,the similar user discovery algorithm based on domain knowledge map is studies,and offline calculation method is adopted to improve efficiency.A formalization recommendation method of multi-user analytical hierarchy process decision-making is used to make decision recommendations,improving user satisfaction and recommendation effects.The experimental verification and analysis of actual health medical scenario data of stroke patients were carried out to prove the predictive accuracy and effectiveness of proposed hybrid recommendation algorithm.